What is Vercel and how does it work?

It watches your repository, builds on every push, and publishes each build as its own immutable deployment behind a unique URL, with static assets served from a global edge network and server code running as functions invoked per request. Every pull request gets a preview URL automatically, which changes client review from a description into a link they can open.

The shape it handles worst is anything long-running. Functions have execution limits, so scheduled jobs, queue workers and heavy background processing need to live somewhere else. Because each invocation is independent it also opens database connections aggressively, which is what a pooler exists to absorb. And usage-based billing means a traffic spike or an inefficient page has a bill attached, worth modelling before launch.